Author Topic: Using MB slows "preparing geometry" - too slow for production  (Read 1138 times)

2019-07-09, 14:54:08

3dkobi

  • Active Users
  • **
  • Posts: 78
    • View Profile
Hi,
I have a simple scene - A cloner clones 2 objects on a distorted sphere. I also have a shader effector which effects the objects size & color.
I have 20,000 clones.
When rendering without object MB - the render is very fast, preparation time is about 20 seconds give or take.
When object MB is activated, the preparation geometry phase is very very very long - I stopped the calculation after 7 minuets.
I'm on threadtripper 1950X.
Scene file is attached. I made it small so it would fit here - to render it like on my machine just change the number of clones to 20,000 and cache the cloner.

Is that a normal behavior ?

2019-07-09, 18:57:30
Reply #1

burnin

  • Active Users
  • **
  • Posts: 835
    • View Profile
Well, opening your scene, following instructions, nothing happens & renders black (R18).
It seems like smthng else is wrong with your scene. Maybe geo or shader deform...? What does AR or Phy show?

Here's a quickie with 20K mograph cubes defective & moblured ;)

2019-07-09, 19:12:43
Reply #2

3dkobi

  • Active Users
  • **
  • Posts: 78
    • View Profile
I'm on R20 so I guess that's the reason it renders black.
I'm rendering this scene now and it renders fine, but without MB....

2019-07-09, 21:25:18
Reply #3

burnin

  • Active Users
  • **
  • Posts: 835
    • View Profile
Ah, ok... and well, takes time to process ~62M effected and blurred tris

here are stats for frame 7 with object motion blur
parsing (single core): 2 min 09 sec
geometry (35% CPU): 3 min 54 sec
12 passes: 40 sec
Total: 6 min 50 sec

2019-07-10, 06:17:44
Reply #4

3dkobi

  • Active Users
  • **
  • Posts: 78
    • View Profile
The result you get is very strange and also wrong. The objects are not moving so fast at all....
Also, in your render, it seems like corona rendered the first frame and over it there's another frame that is blurred much much too strong. At frame 7 the object can't be intersecting because of the dynamic body tag on the cloner....

And another thing - As far as I know, MB is an effect that is calculated during rendering, not during prepration...
On my machine (threadtripper 1950X), for frame 5, I get parsing time of 27 seconds (you got over 2 min), and I get 5 minuets 40 sec "geometry" preparation time....

Any developer here who can look into this ?? I don't think such a long preparation time is normal behavior


2019-07-10, 10:08:08
Reply #5

burnin

  • Active Users
  • **
  • Posts: 835
    • View Profile
A-ok... most likely the difference occured, because i had only baked a single frame.
On 2nd frame, parsing took 0:56 & geo 8:30
Compared to Physical with m-blur which took 11:01 at half the size. Go figure... ;)

I don't think there's anything wrong, but poor optimizations, rather just set "Cloner: Instance Mode" from Instance to Render Instance!
Then you could also disable denoising, at 4% noise level there's really no need  for it with simple motion graphic animations. Likely disable UHD cache, optimize geo... and so on. It's little things, that add up.

2019-07-10, 10:13:32
Reply #6

3dkobi

  • Active Users
  • **
  • Posts: 78
    • View Profile